Morning Brief: Monday, July 20

Sixty-five feeds. Two weeks. 4,170 items reduced to what follows. (what we track, how we crawl, subscribe)

Monday morning. HN front page: Claude Fable produced a counterexample to the Jacobian Conjecture — an unresolved 80-year-old algebraic geometry problem, credit going to an AI system. Companion security signal on the same wire: a researcher found a WordPress RCE (the kind exploit brokers pay $500k for) using GPT-5.6 and about $25 of compute. Cloudflare's Friday WAF post on two high-severity WordPress vulnerabilities now reads as the exact ambient hazard the WordPress-RCE story sits inside. The Monday arXiv drop lands 168 items with the governance and agent-audit cluster extending through the weekend gap: A Methodology for Auditable Trustworthiness Levels in AI Lifecycle Governance, Harmonizing AI Safety Thresholds, Coercion and Deception in AI-to-AI Management, Alipay-PIBench for coding-agent payment integration, Beyond Success Rate: Cost-Aware Evaluation of Offensive and Defensive Security Agents, and Do Coding Agents Need Executable World Models, Simplification, and Verification to Solve ARC-AGI-3? Slashdot has Greg Kroah-Hartman on Rust in the kernel, a free-speech piece on chatbots that won't criticize leaders from repressive regimes, and an SV-tech-workers financial-insecurity story. MIT Technology Review reports AI is more likely than humans to form hiring biases. Willison publishes a Sam Altman quote. Nikita Tonsky posts Looking for work. Leeham: Pratt & Whitney adds CMC composite blades to the next GTF engine.

Top (5-7 min)

Claude Fable produced a counterexample to the Jacobian Conjecture: Hacker News, 2026-07-20. Front-page HN this morning. Alpoge reports Anthropic's Claude Fable produced a counterexample to the Jacobian Conjecture — an 80-year-old unresolved problem in algebraic geometry. Read this first.
Exploit brokers pay $500k for WordPress RCEs. I found one with GPT-5.6 and $25: Hacker News, 2026-07-20. SLCyber researcher walks through finding a WordPress remote code execution bug with GPT-5.6 for roughly $25 in tokens. Reads directly against Friday's Cloudflare WAF disclosure (see Themes).
Rust Will Help Linux Succeed and Makes Coding Fun, Says Greg Kroah-Hartman: Slashdot, 2026-07-20. Kroah-Hartman on Rust-for-Linux. Extends the weekend's Bun-in-Rust claude-code disclosure into the kernel-side Rust adoption arc.
AI is more likely than humans to form biases when hiring: MIT Technology Review, 2026-07-20. MIT TR reports a study finding LLM-based hiring screens exhibit higher measurable bias than human screens. Pairs with today's arXiv Analysing Moral Bias in Finetuned LLMs through Mechanistic Interpretability (see Scan).
A Methodology for Auditable Trustworthiness Levels in AI Lifecycle Governance: arXiv, 2026-07-20. Monday arXiv drop, headline governance paper. Sits with Harmonizing AI Safety Thresholds and Closing the AI Trust Gap: The Case for Independent Certification in a single Monday cluster.
Do Coding Agents Need Executable World Models, Simplification, and Verification to Solve ARC-AGI-3?: arXiv, 2026-07-20. Direct methodological question for coding-agent architectures. Reads with Alipay-PIBench: A Realistic Payment Integration Benchmark for Coding Agents (also today).
Beyond Success Rate: Cost-Aware Evaluation of Offensive and Defensive Security Agents: arXiv, 2026-07-20. Cost-aware evaluation methodology for security agents. Frames the $25/$500k asymmetry the WordPress-RCE story (Top #2) makes concrete.
Quoting Sam Altman: Simon Willison, 2026-07-20. Willison's Monday quote post. Short but on Willison's editorial radar today.
Looking for work: Planet Clojure, 2026-07-20. Nikita Tonsky (Rum, DataScript, Datalevin) posts a job-search post. Notable movement in the Clojure ecosystem.

Themes this week

AI + mathematics breakthrough (Mon): HN / Alpoge: Claude Fable produced a counterexample to the Jacobian Conjecture (Mon). No independent confirmation yet on the wire; worth watching for Willison/Latent Space coverage tomorrow.
AI-assisted vulnerability discovery, live signal: SLCyber: WordPress RCE found with GPT-5.6 for $25 (Mon), Cloudflare: WAF protects WordPress from two high-severity vulnerabilities (Fri), arXiv: Beyond Success Rate — Cost-Aware Evaluation of Offensive and Defensive Security Agents (Mon), arXiv: Evaluating Open-Weight LLMs for Generating Structured Threat Information for Autonomous Vehicle Vulnerabilities (Mon), arXiv: Latent Fusion Jailbreak — Blending Harmful and Harmless Representations to Elicit Unsafe LLM Outputs (Mon), arXiv: Jailbreak Foundry — From Papers to Runnable Attacks for Reproducible Benchmarking (Mon).
Agent governance & audit (Monday arXiv drop): arXiv: A Methodology for Auditable Trustworthiness Levels in AI Lifecycle Governance (Mon), arXiv: Harmonizing AI Safety Thresholds (Mon), arXiv: Closing the AI Trust Gap — The Case for Independent Certification for Trustworthy AI (Mon), arXiv: A Critical Analysis of Trustworthy AI Tools, Mark Frameworks, and the Implementation Chasms (Mon), arXiv: A Formally Grounded ODRL Evaluator (Mon), arXiv: Coercion and Deception in AI-to-AI Management — An Agentic Benchmark of Unprompted Escalation (Mon), arXiv: Causal-Audit — Explicit and Auditable Graph-based Reasoning via Target-Aware Causal Chain Construction (Mon), arXiv: Digital Pantheon — Simulating and Auditing Coalition Formation with LLM Agents (Mon), Alignment Forum: A Red Line and Oversight Framework for Government AI Contracts (Sat, carryover), OpenAI: A scorecard for the AI age (Fri, carryover).
Coding-agent capability & benchmarks (Monday arXiv): arXiv: Do Coding Agents Need Executable World Models, Simplification, and Verification to Solve ARC-AGI-3? (Mon), arXiv: Alipay-PIBench — A Realistic Payment Integration Benchmark for Coding Agents (Mon), arXiv: Evaluating LLM-Based 0-to-1 Software Generation in End-to-End CLI Tool Scenarios (Mon), arXiv: Behavioral Controllability of Agentic Models for Information Extraction — From Fixed Workflows to Reflective Agents (Mon), Willison: Claude Code uses Bun written in Rust now (Sun, carryover), claude-code v2.1.215 (Sun, carryover).
Rust arc (extending Sun's Bun-in-Rust disclosure): Slashdot: Rust Will Help Linux Succeed and Makes Coding Fun, Says Greg Kroah-Hartman (Mon), Willison: Claude Code uses Bun written in Rust now (Sun, carryover).
LLM bias / truth (Mon): MIT Tech Review: AI is more likely than humans to form biases when hiring (Mon), arXiv: Analysing Moral Bias in Finetuned LLMs through Mechanistic Interpretability (Mon), arXiv: Debiasing Text-to-Image Evaluation via Implicit Cultural Alignment Reward Modeling (Mon), Pinboard: Machine Bullshit — Characterizing the Emergent Disregard for Truth in Large Language Models (Mon).
Side-channel & air-gap security: RTL-SDR: TrojPix — Covertly Transmitting Data from Air-Gapped Systems via Video Cable Emissions (Mon), RTL-SDR: War Driving for DECT Devices with a HackRF and Android Device (Mon), Citizen Lab: US Military Smartphones Targeted Through Roaming and Ad Tech (Fri, carryover).
Aviation Monday: Leeham: Pratt & Whitney adding composite materials to next GTF engine (Mon), The Air Current: Boeing quietly scrapped a 777X over rework (Sun, carryover), Leeham: Boeing's opening offer to SPEEA imminent (Sun, carryover).
Clojure / dev culture: Tonsky: Looking for work (Mon), metaredux: Sayid 0.8 (Sat, carryover), metaredux: Stepping Through Macros in CIDER (Fri, carryover).

Scan (15 min)

Monday HN front page
- Claude Fable produced a counterexample to the Jacobian Conjecture, HN, 07-20
- Exploit brokers pay $500k for WordPress RCEs. I found one with GPT-5.6 and $25, HN, 07-20
- Xiaomi-Robotics-1, HN, 07-20
- LoRA Speedrun — a public wall-clock leaderboard for fine-tuning techniques, HN, 07-20
- Moonshine — stream games from your PC to any Moonlight device, HN, 07-20
- How proprietary formats have become Microsoft's main tool for lock-in, HN, 07-20
- US gas prices hit an average of $4 a gallon again, HN, 07-20
Slashdot Monday
- Rust Will Help Linux Succeed and Makes Coding Fun, Says Greg Kroah-Hartman, Slashdot, 07-20
- New Free Speech Concern: When AI Chatbots Won't Criticize Leaders from Repressive Regimes, Slashdot, 07-20
- As AI Transforms Silicon Valley, Some Tech Workers Face Evaporating Financial Security, Slashdot, 07-20
- Zilog Z80 8-Bit CPU Turns 50 — open-source replacement heads to drop-in DIP40 silicon, Slashdot, 07-20
arXiv cs.AI Monday drop (168 items — selected)
- A Methodology for Auditable Trustworthiness Levels in AI Lifecycle Governance, arXiv, 07-20
- Harmonizing AI Safety Thresholds, arXiv, 07-20
- Closing the AI Trust Gap: The Case for Independent Certification for Trustworthy AI, arXiv, 07-20
- A Formally Grounded ODRL Evaluator: Implementation and Comparison, arXiv, 07-20
- A Critical Analysis of Trustworthy AI Tools, Mark Frameworks, and the Implementation Chasms, arXiv, 07-20
- Coercion and Deception in AI-to-AI Management: An Agentic Benchmark of Unprompted Escalation, arXiv, 07-20
- Causal-Audit: Explicit and Auditable Graph-based Reasoning via Target-Aware Causal Chain Construction, arXiv, 07-20
- Digital Pantheon: Simulating and Auditing Coalition Formation with LLM Agents, arXiv, 07-20
- Do Coding Agents Need Executable World Models, Simplification, and Verification to Solve ARC-AGI-3?, arXiv, 07-20
- Alipay-PIBench: A Realistic Payment Integration Benchmark for Coding Agents, arXiv, 07-20
- Evaluating LLM-Based 0-to-1 Software Generation in End-to-End CLI Tool Scenarios, arXiv, 07-20
- Beyond Success Rate: Cost-Aware Evaluation of Offensive and Defensive Security Agents, arXiv, 07-20
- Evaluating Open-Weight LLMs for Generating Structured Threat Information for Autonomous Vehicle Vulnerabilities, arXiv, 07-20
- Latent Fusion Jailbreak: Blending Harmful and Harmless Representations to Elicit Unsafe LLM Outputs, arXiv, 07-20
- Analysing Moral Bias in Finetuned LLMs through Mechanistic Interpretability, arXiv, 07-20
- Debiasing Text-to-Image Evaluation via Implicit Cultural Alignment Reward Modeling, arXiv, 07-20
- BrainPilot: Automating Brain Discovery with Agentic Research, arXiv, 07-20
- Cura 1T: Specialized Model for Agentic Healthcare, arXiv, 07-20
- DSWorld: A Data Science World Model for Efficient Autonomous Agents, arXiv, 07-20
- LLM-Powered Agentic AI for 5G/6G Networks: A Tutorial and Survey on Architectures, Protocols, and Standardization, arXiv, 07-20
RTL-SDR Monday
- TrojPix: Covertly Transmitting Data from Air-Gapped Systems via Video Cable Emissions, RTL-SDR, 07-20
- DeepSDR: Building a Live Public Safety Incident Map with an RTL-SDR, Whisper and an LLM, RTL-SDR, 07-20
- Dragon Labs CR-8: An 8-Channel Coherent SDR Crowdfunding Soon, RTL-SDR, 07-20
- War Driving for DECT Devices with a HackRF and Android Device, RTL-SDR, 07-20
Aviation Monday
- Pratt & Whitney adding composite materials to next GTF engine, Leeham, 07-20
Hackaday Monday
- Could Reticulum Power a Post-Internet Network?, Hackaday, 07-20
- Seven Ways to Install Magnets Into Your 3D Prints, Hackaday, 07-20
- Pong-like Cabinet for Classic Dinosaur Jumper, Hackaday, 07-20
Reading list (Pinboard)
- Machine Bullshit: Characterizing the Emergent Disregard for Truth in Large Language Models, Pinboard, 07-20
- The Attention-Span Class Divide, Pinboard, 07-20
- Just Ask the Dinosaurs How Bad Air Quality Can Get, Pinboard, 07-20
- ribbit/src/rsc.scm — udem-dlteam/ribbit, Pinboard, 07-20
Yesterday's carryover (Sunday, still framing)
- Claude Code uses Bun written in Rust now, Simon Willison, 07-19
- claude-code v2.1.215, claude-code-releases, 07-19
- Qwen3.8 is launching and going open-weight soon, HN, 07-19
- Grok Build open-sourced with zero-data-retention pledge, Slashdot, 07-19
- OpenAI acknowledges GPT-5.6 may accidentally delete files, Slashdot, 07-19
- Boeing quietly scrapped a 777X over rework, The Air Current, 07-19

Tail

The Claude Fable produced a counterexample to the Jacobian Conjecture item is an unverified single-source Twitter/xcancel claim routed through HN. If confirmed, this is the biggest AI-mathematics story of the year to date. Track for Willison/Latent Space coverage tomorrow; if either treats it as real, it moves from "watch" to "top of week".
The WordPress-RCE-for-$25 story concretizes the economics arXiv's Beyond Success Rate: Cost-Aware Evaluation of Offensive and Defensive Security Agents formalises. Cost asymmetry (attacker $25 in tokens vs. broker payout $500k) is the exact metric the paper is proposing to formalise, published on the same day.
arxiv-cs-ai = 168 items today, a normal Monday drop after the Sat/Sun quiet window. Governance/audit heavy: Auditable Trustworthiness Levels, Harmonizing AI Safety Thresholds, Closing the AI Trust Gap, A Formally Grounded ODRL Evaluator, Digital Pantheon, Coercion and Deception in AI-to-AI Management, Causal-Audit. This is the continuation of Wed/Thu's Oracle-Agent-Memory + CAVA + AgentCompass cluster, not a new theme.
bitsavers (6 feeds): connected, 0 items on Monday (sparse archive, expected).
Nikita Tonsky (Rum, DataScript, Datalevin) posting Looking for work is notable ecosystem movement — a well-known Clojure library maintainer publicly job-searching. Not directly load-bearing for the agentic thread, but the Planet Clojure fanout will pick it up.

Feed silences (diagnostic)

Anthropic first-party, Cloudflare, GitHub Blog, GitHub Engineering, Interconnects, Kenneth Payne, Martin Fowler, Microsoft Research, Charity Majors, Quanta Magazine, Grafana Labs, Pluralistic, Andrej Bauer, Databricks, Hugging Face Blog, OpenAI, Nature Machine Intelligence, Hillel Wayne, Schneier on Security, Apple ML Research, Google Research, EFF Deeplinks, wal.sh site, Vercel, LWN, Terence Tao, Tailscale, Netflix Tech Blog, Citizen Lab, FreeBSD Foundation, Latent Space, 404 Media, TechCrunch, Cursor Blog, Pydantic, Alignment Forum, claude-code-releases: no fresh Monday-morning posts yet (Monday publishing tends to land mid-morning to afternoon; expect Latent Space AINews, Willison late-day posts, and any first-party vendor drops to arrive in the afternoon crawl).
James Bornholt: DNS/TLS errors continue (unchanged from prior days).

Build provenance

build: 2026-07-20 | crawler-sha: 5fe7ab8 (Walsh-Research/1.2, compliance v1.3) | feeds: 65 active (78 configured, incl. 12 corp-eng, 6 bitsavers, 5 generated) | items-considered: 4170 (14d, incl. 2282 arXiv) | warehouse: 26543 items | published: 9 | note: Monday morning. HN front page: /Claude Fable produced a counterexample to the Jacobian Conjecture/ (unverified single-source Twitter claim — track for Willison/Latent Space follow-up). Companion security story: WordPress RCE (worth $500k on exploit brokers) found with GPT-5.6 for $25, sitting against Cloudflare's Friday WAF disclosure. Monday arXiv drop: 168 items, governance/audit cluster (Auditable Trustworthiness Levels, Harmonizing AI Safety Thresholds, Closing the AI Trust Gap, Coercion and Deception in AI-to-AI Management, Digital Pantheon, Causal-Audit) extending Wed/Thu's Oracle-Agent-Memory + CAVA + AgentCompass framing. Coding-agent papers: ARC-AGI-3 executable world models, Alipay-PIBench, 0-to-1 CLI tool generation. Security-agent papers: Cost-Aware Offensive/Defensive Security Agents, Latent Fusion Jailbreak, Jailbreak Foundry, AV-vulnerability threat info. Slashdot: Greg KH on Rust in Linux (extends Sun's Bun-in-Rust arc); chatbot-repressive-regime free-speech piece; SV tech workers evaporating financial security; Z80 turns 50 with open-source DIP40 replacement. MIT TR: AI more likely than humans to form hiring biases. Simon Willison: quoting Sam Altman. Planet Clojure: Tonsky posts /Looking for work/. Aviation: Leeham on P&W adding CMC composite blades to next GTF. RTL-SDR: TrojPix air-gap exfil via video cable emissions; DeepSDR (RTL-SDR + Whisper + LLM) public safety map.